We have computed the power to detect linkage under various two-locus models of inheritance in pedigrees and nuclear families and compared methods of linkage analysis. Linkage can generally be detected for epistatic models i the linked locus is dominant and the unlinked locus either dominant or recessive. For an additive model, linkage is more difficult to detect. Th power to detect linkage when the true model is two-locus, is not substantially better when the data are analyzed under a two-locus model vs. a single locus model although the estimates of the recombination fraction are better. If a sample of 200 nuclear families are to be collected for a linkage study, the average sibship size should be 4 with at least 2 affecte sibs, in order to have enough power if inheritance involves two-locus heterogeneity. Pedigree data on linkage of Alzheimer's disease to chromosome 21 and 19 markers are being analzyed as part of the Genetic Analysis Workshop 8 . W are attempting to better localize the gene on chromosome 21 that causes early onset Alzheimer's in some families, by carrying out linkage analysis under more complex models of age-dependent penetrance. Efficiency of linkage calculations for the bipolar mapping project are improving by the development of methods for reducing the number of alleles for marker systems with many alleles and by implementing linkage procedures on high speed computers.